Developing and deploying a use-inspired metapopulation framework for stratified health outcomes

Winter Simulation Conference, 2025

Presenter: Arindam Fadikar

Argonne National Laboratory

2025-12-08

People involved

Roadmap

  • Motivation & collaboration context
  • MetaRVM modeling framework
    • Metapopulation structure
    • Disease progression model
    • Mixing matrices from a synthetic population
  • Implementation as an R package
  • Trajectory-oriented optimization
  • Case study: influenza in Chicago

Motivation

  • Public health decision makers increasingly rely on epidemiological models to:
    • Forecast likely futures
    • Stress-test interventions
    • Plan resource allocation (e.g., beds, staffing)

Motivation

Population-based ODE model

  • Aggregate-level dynamics
  • Homogeneous mixing assumptions
  • Computationally light and fast

Agent-based model

  • Individual-level interactions
  • Heterogeneous contact network
  • Expensive to run for large population

Motivation

Population-based ODE model

  • Aggregate-level dynamics
  • Homogeneous mixing assumptions

MetaRVM

  • Stratified metapopulation structure
  • Time-varying mixing from synthetic contacts

Agent-based model

  • Individual-level interactions
  • Heterogeneous contact network

MetaRVM

  • An open-source R package for modeling infectious disease spread across stratified subpopulations.
  • Subpopulations can be defined by:
    • Geography (e.g., neighborhood, ZIP, zone)
    • Demographics (e.g., age, race/ethnicity)
  • Features:
    • Time varying mixing pattern (weekday vs weekend, day vs night)
    • Extended SEIR-type disease progression
    • Checkpointing functionality
    • Integration with Trajectory-Oriented Optimization workflows

MetaRVM - Disease model

MetaRVM - Population mixing

  • Subpopulations based on age: children (green boxes), working-age adults (blue boxes), and seniors (orange boxes).
  • Different mixing patterns for daytime and nighttime.

Mixing and force of infection

  • For each stratum, the force of infection depends on:
    • Mixing rates between strata \(M = (m_{ij})\)
    • Prevalence of infectious individuals in \(j\)th stratum \(I_t^{(j)}\)

Mixing and force of infection

  • Effective interacting population (exclude H, D)

\[ \mathrm{MP}_t^{(j)} = P_t^{(j)} - H_t^{(j)} - D_t^{(j)}, \qquad j \in \mathcal{J}. \]

  • Mixing across demographic strata

\(M\): mixing matrix \[ M = (m_{ij}), \qquad \sum_{j \in \mathcal{J}} m_{ij} = 1 \ \forall i, \]

\[ \mathrm{MP}_{t,\text{eff}}^{(j)} = \sum_{i \in \mathcal{J}} m_{ij}\,\mathrm{MP}_t^{(i)}, \quad S_{t,\text{eff}}^{(j)} = \sum_{i \in \mathcal{J}} m_{ij}\,S_t^{(i)}, \quad I_{t,\text{eff}}^{(j)} = \sum_{i \in \mathcal{J}} m_{ij}\,I_t^{(i)}. \]

Mixing and force of infection

  • Force of infection (susceptible vs vaccinated)

\[ \lambda_{s,t}^{(j)} = \beta_s \frac{I_{t,\text{eff}}^{(j)}}{\mathrm{MP}_{t,\text{eff}}^{(j)}}, \qquad \lambda_{v,t}^{(j)} = \beta_v \frac{I_{t,\text{eff}}^{(j)}}{\mathrm{MP}_{t,\text{eff}}^{(j)}}. \]

  • S → E transition in stratum (j)

\[ p_{\mathrm{SE}}^{(j)} = 1 - \exp\!\big(-\lambda_{s,t}^{(j)} \,\Delta t\big), \\ \Delta SE_t^{(j)} = p_{\mathrm{SE}}^{(j)}\,S_t^{(j)}. \]

Age-stratified model - illustration

What we need

  • Population counts
    • Source: Census
  • Daily vaccination schedule
    • Source: CDPH
  • Mixing matrices
    • Source: ChiSIM synthetic population

Age-stratified model - illustration

Synthetic population

  • Statistically representative synthetic population of Chicago
    (2.7M people, 1.4M places, 13k activity schedules)

  • Demographically accurate households from ACS + PUMS at the CBG level

  • Workplaces generated from County Business Patterns + LEHD OD data
    (individuals assigned to realistic work locations)

  • Schools assigned to all school-aged children
    (some adults assigned schools as workplaces)

  • Additional mixing locations (restaurants, gyms, etc.) from SafeGraph

Age-stratified model - illustration

Mobility –> Mixing

  • Agent mobility from real schedules

    • Adults: American Time Use Survey
    • Children: Panel Study of Income Dynamics
    • Each agent draws 1 of 10 possible schedules daily (weekday/weekend)
  • Endogenous contact network emerges from hour-by-hour co-location of agents

  • Aggregate

    • Count the total number of contacts that occur at each location between and within subpopulations
    • Normalize the counts to arrive at per-capita value

Daytime vs Nightime mixing

Daytime

  • Strong location based mixing

Nighttime

  • More household and community mixing

Vaccination

TOO: overview

  • Calibration goal:
    • Align model-generated hospitalizations with observed time series
  • Approach:
    • Choose a set of calibration parameters (e.g., transmission scaling, reporting/observation fraction)
    • Define a loss function (e.g., squared error between modeled and observed hospitalizations)
    • Use Bayesian optimization to:
      • Explore parameter space efficiently
      • Balance exploitation and exploration
  • Outcome:
    • Posterior-like distribution over plausible parameter values
    • Uncertainty envelopes around model trajectories

Results

Summary

  • MetaRVM provides a use-inspired metapopulation modeling framework for:
    • Detailed, stratified tracking of health outcomes
    • Infectious disease spread across multiple interacting subpopulations
  • Key contributions:
    • Uses synthetic population-derived mixing matrices
    • Balances realism with computational efficiency
    • Delivered as an open-source R package with a Shiny front end
  • Case study:
    • Demonstrated on influenza-related hospitalizations in Chicago
    • Supports prospective surveillance and scenario analysis

Funding acknowledgments

  • Chicago Department of Public Health

Questions?

Thank you!